Computing multi-valued physical observables for the high frequency limit of symmetric hyperbolic systems

Shi Jin, Hailiang Liu, Stanley Osher, Richard Tzong‐Han TsaiJournal of Computational Physics 2005 · DOI

TL;DR

This paper introduces a novel level set method to accurately compute multi-valued physical observables (like density and velocity) for high-frequency wave propagation in symmetric hyperbolic systems. By leveraging phase space and Liouville equations, it overcomes limitations of traditional methods, enabling more robust simulations of complex wave phenomena.

What Problem Does This Paper Address?

Simulating high-frequency wave propagation, especially in complex media, is computationally challenging. Traditional methods often struggle with the "multi-valued" nature of physical observables that arises when waves cross or interact. This multi-valuedness, where a single point in space might have multiple values for density or velocity, leads to singularities in standard mathematical descriptions. The paper specifically addresses the "high-frequency limit" of symmetric hyperbolic systems, which is crucial for understanding phenomena like geometric optics and the semiclassical limit of quantum mechanics. Existing approaches, like those based on the standard kinetic equation with Dirac measures, often suffer from numerical oscillations and smearing. The authors aim to develop a method that can accurately capture these multi-valued solutions without these drawbacks.

Methods

The paper proposes two main approaches, both rooted in the concept of evolving quantities in phase space (position and momentum/wave vector).

  1. Decoupled Eikonal and Transport Equations: This method starts by solving a decoupled system: an eikonal equation for the wave's phase ($S$) and a transport equation for its density ($p$). The core idea is to track the density evolution near the "bi-characteristic manifold" in phase space. This manifold is defined by the common zeros of several level set functions. These level set functions are generated by solving the Liouville equation, which describes the evolution of probability densities in phase space. A key innovation is the introduction of a new quantity $f$ that, when integrated along the bi-characteristic manifold, resolves the multi-valued density and higher moments.
  2. Wigner Transform Approach: This second approach leverages the high-frequency limit derived from the Wigner transform of the original symmetric hyperbolic system. This naturally leads to Liouville equations in phase space, but with measure-valued initial data. The linearity of the Liouville equation allows for a decomposition of the density distribution, enabling the computation of higher-order moments like energy and energy flux.

A significant advantage highlighted is the use of $L^\infty$ initial data for the Liouville equations, avoiding the need to directly handle Dirac measures. Furthermore, a local level set method is employed to reduce computational costs in the high-dimensional phase space.

Datasets

Not discussed in the paper. The paper focuses on developing a general numerical method and demonstrates its application to specific wave equations (optical, acoustic, and general symmetric hyperbolic systems) rather than evaluating it on pre-existing datasets.

Key Insights and Ideas

  • Phase Space as the Key: The paper strongly emphasizes that understanding and simulating multi-valued wave phenomena requires working in phase space (position and momentum/wave vector). This allows for a more natural representation of wave characteristics and their evolution.
  • Level Sets for Multi-Valuedness: Level set functions are ingeniously used to implicitly define and track the multi-valued nature of observables by identifying the zero level sets of these functions in phase space.
  • Liouville Equation as the Workhorse: The Liouville equation proves to be a central tool for evolving quantities in phase space, whether it's the phase gradient, density, or coherence matrices.
  • Avoiding Dirac Measures: A crucial insight is that by reformulating the problem in phase space and introducing auxiliary quantities like $f$, one can avoid the numerical difficulties associated with directly handling Dirac delta functions, which are common in traditional high-frequency limit methods.
  • Computational Efficiency: The use of local level set methods is highlighted as a way to significantly reduce the computational burden in the high-dimensional phase space.

Key Results

The paper presents a framework and methodology for computing multi-valued physical observables. While specific quantitative results are demonstrated through numerical examples (Figures 5.1-5.9), the paper's primary contribution is the development of the methodology itself. The numerical examples illustrate:

  • The successful computation of self-crossing wavefronts for optical waves with constant and variable speeds.
  • The application to wave guides, showing energy transport.
  • The handling of contracting circles and ellipses in 2D, demonstrating the method's ability to capture complex evolving wavefronts.
  • The method's applicability to acoustic waves.

The paper demonstrates that the proposed level set approach can accurately capture the complex, multi-valued behavior of wave propagation without the oscillations and smearing often seen in other methods.

Limitations and Future Work

  • Singularity at $k=0$: For certain Hamiltonians, like those in the wave and acoustic equations, the Hamiltonian has a singularity at $k=0$. The level set method can only be applied when the space gradient $VS$ stays away from zero. The authors suggest that suitable boundary conditions might be needed at this singularity.
  • Non-continuous $VS$: The paper notes that if the initial phase gradient $VS_0$ is not continuous, the Dirac delta function in initial data becomes ill-defined. This is an open question, and the authors suggest regularization techniques (like embedding the completion of the subgraph) but acknowledge that further research is needed on how to properly regularize WKB initial data in such cases.
  • General Symmetric Hyperbolic Systems with Multiple Eigenvalues: While the paper outlines the approach for systems with multiple eigenvalues, the numerical implementation and detailed results for this more complex case are not fully explored in this paper.
  • Further Numerical Exploration: The paper mentions that numerical techniques are documented in previous works and that further efficiency can be gained from specific numerical treatments (e.g., for delta functions). Future work could involve more extensive numerical studies and optimizations.

Summary updated

BibTeX
@article{jin2005computing,
  author = {Shi Jin and Hailiang Liu and Stanley Osher and Richard Tzong‐Han Tsai},
  title = {{Computing multi-valued physical observables for the high frequency limit of symmetric hyperbolic systems}},
  journal = {Journal of Computational Physics},
  year = {2005},
  doi = {10.1016/j.jcp.2005.04.020},
}
Figure